Finite Elements in Analysis and Design 31 (1998) 114

A finite element analysis for the least temperature

rise in a hot torsion test specimen
M. Zhou*, M.P. Clode
Department of Mechanical Engineering, Kings College, University of London, London WC2R 2LS, UK

Abstract
The temperature distribution in a hot torsion specimen, caused by plastic deformation, is predicted by a finite element
analysis with the aim of obtaining an optimum specimen geometry that would produce the least heat accumulation
during twisting so as to avoid the possible flow localization and to facilitate the interpretation of the data obtained. The
prediction of temperature rise is made for an aluminium alloy 5252 for different twist rate, initial temperature and
specimen geometry, from which a procedure for the least heat generation is suggested. The constitutive equations are
modified for the unavoidable temperature rise and the predictions so obtained are compared with the experimental
data. ( 1998 Elsevier Science B.V. All rights reserved.
Keywords: Finite element analysis; Temperature rise; Hot torsion testing; Flow softening; Aluminium alloys

1. Introduction
In the modelling of hot metal forming processes such as extrusion of aluminium alloys, hot
torsion testing can be used in generating the data of flow stress to large strains to validate the
constitutive equations used. However, heat may be generated during testing due to large plastic
flow which may not be able to dissipate away in a time required to achieve the given deformation.
As a result, temperature may rise and distributions may become inhomogeneous, which may
induce flow localization and softening [1]. The interpretation of the data obtained may also be
affected because the excessive heat generation intensifies flow softening.
The deformation behavior of aluminium alloys at hot working temperatures is primarily
characterized by dynamic recovery [2,3]. A steady state or saturated flow stress is obtained as
a result of the annihilation of pairs of dislocations and the development of an equiaxed subgrain

* Corresponding author. Tel.: #44 01709 825159; e-mail: [email protected]

0168-874X/98/$ see front matter ( 1998 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 8 - 8 7 4 X ( 9 8 ) 0 0 0 4 4 - 4

M. Zhou, M.P. Clode / Finite Elements in Analysis and Design 31 (1998) 114

structure. However, during hot torsion testing, the flow stress curve can also be observed to decline
after its peak value has been obtained. This flow softening has been attributed to the internal heat
generation due to large plastic deformation [4]. To simplify the analysis, isothermal or adiabatic
assumptions may be adopted, but these assumptions are only valid, respectively, for very low or
very high deformation rate [5,6]. In the analysis related to flow stresses over a strain rate range
where conditions are neither isothermal nor adiabatic, the omission of temperature changes can
invalidate the derived constitutive equations.
Temperature changes during torsion testing depend on the twist rate applied, initial temperature, angle of twist, and the geometry of specimen. The geometry of specimen, such as the
diameter and length of the specimen gauge and head, has evolved based primarily on experience
and test machine limitations. Inappropriate design of the test specimen may lead to a high
accumulation of heat, caused by large plastic deformation, in the central region of the specimen
gauge length.
The variation of temperature distribution with the initial temperature, twist rate and the
geometry of specimen due to plastic deformation is investigated in this paper by a finite element
analysis with the aim of obtaining an optimum specimen geometry that would produce the least
heat accumulation during twisting. The finite element model is formulated for the twisting
specimen which is heated by an induction coil and controlled by a thermocouple. Account is taken
of the temperature rise from plastic deformation and the consequential heat transfer by conduction
to the non-deforming specimen ends and by convection and radiation to the environment. The
predictions of the temperature rise for an aluminium alloy 5252 are discussed. An optimization
procedure for the specimen geometry is suggested based on the consideration of stiffness requirement and the least heat generation. The constitutive equations are modified for the unavoidable
temperature rise and the predictions of the stressstrain curves so obtained are compared with the
data of aluminium alloy 5252.

2. A heat conduction model in hot torsion testing

A solid specimen with a geometric shape of stepped cylinder is considered in the present work
and shown in Fig. 1. The gauge section of the specimen is designed to be relatively thin which has
a length of and a diameter of D . Both ends of the specimen have a larger diameter of D . The
'
'
)
gauge part of the specimen is deformable during testing and the ends are regarded as rigid. An
induction coil is used to heat the specimen to the required initial temperature which is then held for
2 min to ensure that temperature distribution is as uniform as possible. The specimen is twisted on
a high-temperature deformation simulator in torsional mode at different twist rates. It can be seen
that heat which is generated from the plastic work in the gauge section of the specimen will transfer
by conduction to the non-deforming specimen ends and by boundary convection and radiation to
the surrounding environment.
The relationship between the sizes of the specimen, as shown in Fig. 1, is chosen to meet the basic
stiffness requirement expressed approximately as
k
'"j or
k
)

A BA B

D 4
'
) "j,
D

)
'

(1)

M. Zhou, M.P. Clode / Finite Elements in Analysis and Design 31 (1998) 114

Fig. 1. The geometry of a solid hot torsion specimen.

Fig. 2. (a) A model of finite element analysis for heat conduction and (b) the distribution of initial temperatures.

where k and k are the torsional spring rates [7] for the gauge and head of the specimen,
'
)
respectively, and j is a ratio less than 1.0 indicating a smaller torsional spring stiffness in the gauge
than in the head.
As the constraints of specimen ends and temperature environment are symmetric with respect to
its centroidal axis, it is assumed that the heat conduction in the specimen is independent of the
circumferential angle of twist and can therefore be simplified by a 2D idealization. The resulting
model is shown in Fig. 2a where only one-quarter of a longitudinal section is considered because of
its mirror symmetry with respect to r- and z- axis.

M. Zhou, M.P. Clode / Finite Elements in Analysis and Design 31 (1998) 114

The axisymmetric conditions imply that heat flux along the boundaries S and S are zero,
1
2
or insulated. As the boundary S is directly heated by the induction coil which is controlled by
3
a thermocouple marked as S in Fig. 1, a temperature that is not lower than the initial temperature
is maintained throughout the specimen. The boundary S is exposed to the surrounding environ4
ment where the energy loss is considered through both boundary convection and radiation.
Fig. 2b shows the distribution of the initial temperature where the temperature at the boundary S is prescribed as and the region covered by the induction coil has an initial temperature
5
B
of .
i
3. FEA formulation for heat conduction
The equations governing the heat conduction of the specimen, as modelled in Fig. 2a, may be
derived from the following functional:

PP C A B A B
A
B D
P C A
A
BD

P"2n

1
L 2 1
L 2
k
# k
2
Lr
2
Lz

! q!oc
#e p
B B

Lt

5
!4
r
5

r dr dz#2n

S3,S4

h
B

2
!
aB
2

r ds,

(2)

with the necessary boundary conditions given by

" (t) on S ,
B
5

L
"0 on S
1
Lr

and

L
"0 on S .
2
Lz

(3)

In these expressions, is temperature, k is thermal conductivity, q is the internal heat generation

per unit time and per unit volume, o is mass density, c is specific heat, t is time, h is convection heat
B
transfer coefficient, is the ambient temperature of the surrounding environment, e is the
aB
B
emissivity of the specimen surface, p is the StefanBoltzman constant, is the surface temperB
r
ature of the furnace, is the temperature prescribed on the boundary S . The material is assumed
B
5
to be homogeneous and isotropic.
The heat source q in Eq. (2) is assumed to be generated by plastic work as follows:
q"cwR ,
(4)
1
where c is the fraction of the plastic work converted into heat during deformation and it is
approximately taken as 0.95, wR is the plastic work per unit volume and per unit time which equals
1
pN eNR with pN being the effective stress and eNR the effective strain rate in the sense of the von Mises yield
criterion. In uniaxial tension, the rate of plastic work wR "peR is obtained with p being the true
1
stress and eR the strain rate. In torsion, the effective stress is J3 times the shear flow stress and the
effective strain rate is 1/J3 times the shear strain rate. Therefore, wR "qcR can be obtained where
1
q is the shear stress and cR the shear strain rate. It is also assumed that twisting is uniform so that
cross-sections of the specimen remain plane and rotate as if rigid during deformation. This

M. Zhou, M.P. Clode / Finite Elements in Analysis and Design 31 (1998) 114

assumption results in the following linear shear strain and shear strain rate in terms of radius
coordinate r,
r
c" h and

'

r
cR " hQ

'

(5)

where is the gauge length of the specimen, h and hQ are angle of twist and twist rate, respectively.
'
In the calculation of the plastic work shown in Eq. (4), a constitutive equation that relates stress,
strain, strain rate and temperature is required. For aluminium alloys, the material behavior at hot
working temperatures is characterized by competing hardening and recovery processes [2,3]. The
constitutive model developed by Zhou and Clode [9] may be used to describe the high-temperature deformation properties which exhibit a saturated stress response to large deformation as
a result of dynamic recovery. In uniaxial case, the model can be expressed as,

p"C[1!exp(!ben)]sinh~1 BeR m exp

A BD
mQ
R

(6)

where is the temperature, R is the universal gas constant and Q is the activation energy. The
material constants, C, B, m and Q, are determined by the saturated flow stresses for different strain
rate and temperatures. The two parameters which are strain rate and temperature dependent, b and
n, are defined as

BD

BD

bQ
b"a eR ~bb exp ! b
b
R

(7a)

and
bQ
n"a eR ~bn exp ! n
n
R

(7b)

where a , b , a and b are constants.

b b n
n
In the case of torsion, the effective stress, strain and strain rate can be converted from the torsion
test data at the effective radius [8], r "0.361D , as follows:
%&
'
12J3
r
r
pN "
M, eN " %& h and eNR " %& hQ ,
nD3
J3
J3
'
'
'

(8)

where M is torque, h is angle of twist and hQ is twist rate. Thus, the material constants in Eqs. (6), (7a)
and (7b) can be determined by the measurements of torque as a function of twist, twist rate and
temperature [9].
Considering the conditions for stationary value of the functional in Eq. (2) gives the heat
conduction governing equations as

A B A B

1 L
L
L
L
L
rk
#
k
#cwR "oc
1
r Lr
Lr
Lz
Lz
Lt

(9)

M. Zhou, M.P. Clode / Finite Elements in Analysis and Design 31 (1998) 114

with the natural boundary conditions as

L
L
! k
n #k
n "h (! )#e p (4#4) on S and S ,
B
aB
B B
r
3
4
Lr r
Lz z

(10)

where n and n are the direction cosine of the unit outward normal to the boundaries S and S .
r
z
3
4
Eq. (9) represents the energy balance between heat conduction in the specimen along r and
z directions, internal heat source produced by plastic flow and the time related transient term.
Eq. (10) implies that heat convection and radiation only occur on the boundaries S and S and the
3
4
furnace acts as a blackbody regardless of the emissivity of its surface. If the heat conduction terms
are not considered, Eq. (9) becomes
L
cwR "oc
1
Lt

(11)

which is a locally adiabatic process [5] and is equivalent to the deformation process at high strain
rate. The temperature rise at high deformation rate may therefore be estimated by

c eL
pN (eN , eNR , ) deN ,
(12)
D(eL , eNR , )"
oc
0
where the effective strain eL indicates the deformation dependence of temperature rise and pN (eN , eNR , )
describes the dependence of the effective stress on strain, strain rate and temperature.
Temperature changes during plastic deformation which as a function of the coordinates r and
z may be approximated by a function of the Lagrangian type at an element level as
+M ,Na% on element A%

(13)

where N"MN , N , N , N N is the shape functions defined for 4-node isoparametric elements [10],
1 2 3 4
and (a%)T"M , , , N is the nodal temperature of the element.
1 2 3 4
The approximation assumed in Eq. (13) is substituted into Eq. (2) and the following FE model
on the element level can be obtained by considering the stationary condition:
H%aR %#K%a%"f %,

(14)

where aR % is the time differentiation of a%, H% and K% are, respectively, the element capacitance and
conductance matrixes defined by

PP
PP C A B
P

ocNTNr dr dz,
A%
LN
LNT
LN
LNT
k
#
k
Lr
Lz
Lz
Lr
A%

H e"2n
K%"2n

A BD

(15a)
r dr dz

[h NTN#e p NTN(Na%)3] r ds
B
B B
S%3, S%4
and the nodal force vector on the right-hand side is defined by
#2n

PP

f %"2n

cwR NTr dr dz#2n

1

%
3

S ,S

%
4

[h NT #e p 4NT]r ds.
B
aB
B B r

(15b)

(15c)

M. Zhou, M.P. Clode / Finite Elements in Analysis and Design 31 (1998) 114

The resulting assembly matrix equation, HaR #Ka"F, is nonlinear in terms of the assembly nodal
temperatures a, which may be solved by the following iteration format:
HaR #K(a )a "F (i"1, 2,2),
(16)
i
i~1 i
i
where a (i"1) is the initial nodal temperature of the specimen.
0
Time integration has to be carried out for the time history of nodal temperatures, which may be
obtained from the format given as follows:
[H#bDtK(a )]a "[H!(1!b)DtK(a )]a
i~1 i
i~1 i~1
#[(1!b)F #bF ]Dt (0)b)1).
(17)
i~1
i
The integration time increment must be constrained to ensure that the computation is stable and
accurate, which may be derived by the second law of thermodynamics in the calculation of heat
conduction [11] and given as
oc(Dx)2
DtK
,
4k

(18)

where (Dx) is the minimum dimension of an element.

4. Predictions of temperature distribution

The solid circular cylindrical specimen shown in Fig. 1 is illustrated, which has a gauge length of
10 mm, a gauge diameter of 10 mm and a head diameter of 20 mm. The material tested was an
aluminium alloy 5252 which consists of the primary alloying elements in weight percentage:
magnesium 2.5, manganese 0.05, silicon 0.08, iron 0.21, copper 0.03, titanium 0.014 and boron
0.004. Homogenization was made before testing to reduce residual stresses and to achieve uniform
structure.
A finite element computer program has been written based on the FE model discussed above.
The constitutive equations in Eqs. (6), (7a) and (7b) have been validated by the data of aluminium
alloy 5252. The torque-twist measurements in the tests are converted into the equivalent
stressstrain data at the effective radius [8] and then used to determine the material constants in
the equations [9]. The temperature is assumed to be constant as an initial approximation of the
stress-strain relationship. The material constants obtained include: C"30.441, B"2.940]10~5,
m"0.441, Q"168.650 kJ mol~1, a "109.767, b "0.07534, a "1.68, and b "0.01517.
b
b
n
n
Fig. 3 shows the application of the present FE model to a rod with a uniform heat source per
unit volume and a given temperature at one end and insulated at the other. The temperature
B
distribution is shown to be in a good agreement with the analytical one, which verifies the heat
conduction analysis in the program.
Fig. 4 shows the comparison of temperature rise on the central cross section of the specimen
along the radius axis at an initial temperature of 450C to achieve a surface shear strain of 4.8. As
expected, there is no temperature rise for isothermal condition, indicated by the broken line. For
adiabatic condition, temperature rise is indicated by the lines with dark square. Predicted by
Eq. (12), it can be seen to vary from zero at the center to maximum at the surface and have the

M. Zhou, M.P. Clode / Finite Elements in Analysis and Design 31 (1998) 114

Fig. 3. Temperature distribution in a rod with a uniform heat source and heated at one end (z"0) and insulated at the
other (z").

Fig. 4. Comparison of temperature rise for isothermal, adiabatic and non-isothermal conditions on the central cross
section of the specimen (z"0).

similar distribution of shear stress along the radius axis. The lines with light square show the
results for non-isothermal condition which consider heat transfer to the regions of lower temperature and to the non-deforming specimen heads. It is shown that temperature gradient tends to
decline and temperature distribution tends to be uniform along the radius axis as a result of heat
conduction.

M. Zhou, M.P. Clode / Finite Elements in Analysis and Design 31 (1998) 114

The contour of the temperature rise in the gauge length of the specimen is shown in Fig. 5a and
b which is subjected to a high twist rate of 15.0 rev s~1 and a low twist rate of 1.4 rev s~1,
respectively. In these figures, the gauge axis is referred to the centroidal axis of the specimen
indicated by r"0. It is found that the maximum temperature rise occurs on the surface of the
central cross section because heat generation accumulates near the symmetric axis z"0 and the
maximum deformation occurs on the surface. These results are more prominent for high deformation rate which may become one of the dominant factors for local flow softening of materials.
The above demonstration has shown that temperature varies significantly along the specimen
axis at high twist rate and the maximum temperature rise occurs in the central cross section of the
specimen. Local flow softening of materials may occur if the excessive heat accumulates around
that region. It was also shown that temperature rise is not significant at low strain rate. For
example, a maximum temperature rise of 1C is produced for twist rates less than 0.16 rev s~1. The
reason for this is that slow twisting allows sufficient time for heat dissipating through conduction
from the deformable gauge length to the rigid head and through convection and radiation from the
specimen boundaries to the environment. It is therefore reasonable to adopt an isothermal
assumption in the analysis.

5. Temperature variation with specimen geometry

Temperature rise depends not only on strain rate, strain and initial temperature, as illustrated
above, but also on the geometry of specimen. Fig. 6 shows the distribution of temperature rise
along the specimen axis (r"0) at an initial temperature of 450C to achieve a surface shear strain
of 4.8. The temperature rise for two twist rates, 15.0 and 1.4 rev s~1, is demonstrated. The gauge
diameter is assumed to be a constant of 10 mm and the gauge length to vary from 6 to 14 mm. The
comparison from the figure shows that, if the gauge diameter is fixed, a short gauge length
generates a low temperature rise in the case of high twist rate, which is the result of a faster heat
conduction from the gauge length to the rigid heads of the specimen. In this case, a high strain rate
at the effective radius can be achieved, see Eq. (8). In the case of low twist rate, the geometry of the
specimen does not have much effects on temperature rise.
If the gauge length is fixed to be 12 mm and the twist rate is 15.0 rev s~1 and a surface shear
strain of 4.8 is to be achieved, as shown in Fig. 7, a large gauge diameter will produce a low
temperature rise, which is more significant at low initial temperatures. Meanwhile, a high strain
rate at the effective radius can be obtained.
It has been shown from the above demonstration that an increase in the gauge diameter and
a decrease in the gauge length leads to a decrease in temperature rise. The least temperature rise
design may be achieved in this way but doing so will increase the relative stiffness of the gauge
length. To compromise, an optimization procedure may be followed to obtain a specimen
geometry that not only meets the basic stiffness requirement but also generates heat as less as
possible to avoid flow localization, which is outlined as follows.
(a) For an extruded rod with a diameter of D (or larger), the length of specimen head and one
)
)
of the parameters of the gauge length ( or D ) are chosen, then another parameter of the
'
'
gauge length is determined by Eq. (1) for the given ratio of the torsional spring rates k and k .
'
)

10

M. Zhou, M.P. Clode / Finite Elements in Analysis and Design 31 (1998) 114

Fig. 5. The contour of temperature rise in the gauge length of the specimen at 15.0 rev s~1 (a) and 1.4 rev s~1 (b).

M. Zhou, M.P. Clode / Finite Elements in Analysis and Design 31 (1998) 114

11

Fig. 6. Temperature rise along the specimen axis (r"0) with gauge diameter fixed and gauge length varied.

Fig. 7. Temperature rise along the specimen axis (r"0) with gauge length fixed and gauge diameter varied.

(b) The temperature distribution is estimated for the parameters chosen in Step (a). If the
temperature rise predicted is too high, which may result in flow localization at high rate of
twisting, the parameters of , and D are modified based on the conclusions obtained on the
) '
'
variations of temperature rise with the gauge length and gauge diameter.

12

M. Zhou, M.P. Clode / Finite Elements in Analysis and Design 31 (1998) 114

6. Effects of temperature rise on the stress-strain relationships

Though the design of specimen geometry so obtained may ensure the least heat generation
during twisting, heat may still inevitably be produced in the gauge length of the specimen due to
large plastic deformation, especially at high twist rates. So the constitutive equations that characterize material flow should be modified before they are applied to modelling of metal forming
processes.
It has been demonstrated that the distribution of temperature rise in the gauge length of the
specimen is non-uniform, which varies in both radian and axial directions. An average volume
temperature [4] may be defined in the validation of constitutive equations. However, if the data for
the stressstrain curve is obtained from the conversion of twist-torque measurements at the
effective radius, the average effective radius temperature may be more appropriate, which is defined
as

2 L'@2
( , hQ , h)"
(r , z; , hQ , h) dz
(19)
% 0
%&
0

0
where (r , z; , hQ , h) is the temperature distribution at the effective radius which is a function of
%&
0
initial temperature, twist rate, twist and the specimen geometry and predicted by the FE model in
this paper.
The constitutive equations in Eqs. (6), (7a) and (7b) are then modified by replacing the temperature term with the average effective radius temperature . Fig. 8 shows the predictions of the
%
stress-strain curves of aluminium alloy 5252 at an initial temperature of 500C and at twist rates of
15.0, 5.0, 1.4, 0.16 and 0.015 rev s~1. The corresponding effective strain rates at the effective radius

Fig. 8. The predicted stressstrain curves of aluminium alloy 5252, as indicated by solid lines, in which temperature rise
has been considered.

M. Zhou, M.P. Clode / Finite Elements in Analysis and Design 31 (1998) 114

13

are evaluated from Eq. (8) to be 19.70, 6.44, 1.84, 0.21, and 0.02 rad s~1, respectively. The broken
lines represent the predictions that do not include the effects of temperature rise and the solid lines
are those modified by incorporating the average effective radius temperature. It is shown that
a reasonable prediction of flow stress as a function of strain, strain rate, and instantaneous
temperature has been obtained.

7. Conclusions
As inappropriate design of the test specimen in hot torsion testing may lead to a high
accumulation of heat in the central region of the specimen gauge length, flow localization may
occur and the interpretation of the test data may be affected. To avoid this, knowledge of the
variation of temperature rise due to plastic deformation over the specimen gauge length for
different specimen geometry and loading conditions is necessary. It is also necessary to estimate the
distribution of temperature rise in the event that heat generation is inevitable so that the
constitutive equations used are modified accordingly.
A heat conduction model to predict by finite element analysis the temperature rise during plastic
deformation in hot torsion testing has been discussed. The predictions of temperature changes in
an aluminium alloy 5252 consider the effects of deformation, deformation rate and initial temperature. The effects of specimen geometry on temperature rise caused by the plastic work have also
been investigated.
It has been shown that, at low twist rate and therefore low effective strain rate, temperature rise is
small and temperature distribution tends to be uniform. It is largely governed by twist rate, initial
temperature and angle of twist. The specimen geometry has little effect. At high initial temperature,
the temperature rise and its gradient tend to decline, even at high twist rate. However, at high twist
rate and low initial temperature, the effects of the specimen geometry are significant. An increase in
the gauge length and a decrease in the gauge diameter increase temperature rise. An optimum
geometry procedure may be followed which considers compatibility between the stiffness and the
least temperature rise requirement.
Though the specimen may have been designed with the criterion of the least temperature rise,
heat generation, more or less, is still inevitable. The constitutive equations have to be modified
accordingly. The predictions of the stressstrain curves of aluminium alloy 5252 made by the
modified constitutive equations have been shown to be in a good agreement with the data.

References
[1] S.L. Semiatin, G.D. Lahoti, Deformation and unstable flow in hot torsion of Ti-6Al-2Sn-4Zr-2Mo-0.1Si, Metallurgical Trans. A 12 A (1981) 17051717.
[2] C.M. Sellars, Modeling microstructural development during hot rolling, Mater. Sci. Technol. 6 (1990) 10721081.
[3] M.P. Clode, Hot torsion testing to model the extrusion processes, Aluminium Ind. 11 (1992) 3439.
[4] T. Sheppard, D.S. Wright, Deformation of flow stress: Part 1. constitutive equation for aluminium alloys at elevated
temperatures, Met. Technol. 6 (1979) 215223.
[5] A.U. Sulijoadikusumo, O.W. Dillon Jr., Work softening of Ti-6Al-4V due to adiabatic heating, in: R.W. Rohde et al.
(Eds.), Metallurgical Effects at High Strain Rates, Plenum Press, New York, 1973, pp. 505517.

14

M. Zhou, M.P. Clode / Finite Elements in Analysis and Design 31 (1998) 114

[6] G.R. Johnson, Dynamic analysis of a torsion test specimen including heat conduction and plastic flow, ASME J.
Eng. Mat. Technol. 103 (1981) 201206.
[7] J.E. Shigley, L.D. Mitchell, Mechanical Engineering Design, McGraw-Hill, New York, 1983.
[8] D.R. Barraclough, H.J. Whittaker Jr., K.D. Nair, C.M. Sellars, Effect of specimen geometry on hot torsion test
results for solid and tubular specimens, J. Testing Evaluation 1 (1973) 220228.
[9] M. Zhou, M.P. Clode, A constitutive model and its identification for the deformation characterized by dynamic
recovery, ASME J. Eng. Mat. Technol. 119 (1997) 138142.
[10] R.D. Cook, Concepts and Applications of Finite Element Analysis, Wiley, New York, 1974.
[11] J.P. Holman, Heat Transfer, McGraw-Hill, New York, 1990.